Kern feature writer: Add option to support mark filtering sets

When kerning a base to a mark, one might need to skip over certain irrelevant marks to make the kerning work. Currently there's no way to tell the kern feature writer to set a mark filtering set for base-mark kerning. I don't think this is hard to do but I don't know what a good interface to it would look like. (For example, assuming you have a Glyphs source...)

Start by locating the kern feature writer and reviewing how base-mark kerning is generated, along with the existing handling of Glyphs sources. The issue does not name files or tests, and completion would require agreeing on an interface for specifying mark filtering sets before implementing and testing it.

Written by the indexing model from the issue text.